Ternary PUF-based Secure Mutual Authentication Using RNNs for Sequence Learning in Response Classification
Jiayi Chang, Shekoufeh Neisarian, Nico Mexis, Nikolaos Athanasios Anagnostopoulos, Tolga Arul, Elif Bilge Kavun · 2025
In the Internet of Things (IoT), authentication plays an important role. Physical Unclonable Functions (PUFs) are a cost-efficient, lightweight authentication approach for many devices in IoT. Ternary PUFs are a type of PUF that generates responses containing three different values instead of the usual two for binary response bitstreams, which is more resistant to machine learning attacks. This paper proposes an authentication scheme using ternary PUFs and a sequence learning machine learning model. The sequence learning model could process a PUF’s sequence of “trits” as the response. If the model determines that the response is from a PUF that is authorized for authentication, the server would then authenticate the PUF that sent the response. The model is compared to baseline Convolutional Neural Networks (CNNs). The experimental results demonstrate that sequence learning-based Recurrent Neural Network (RNN) models achieve superior classification accuracy compared to baseline CNNs in authenticating ternary PUF responses.